| With the development of society,energy problems have gradually become prominent,and energy shortages have appeared worldwide.In order to alleviate the pressure of energy,we must start from both the source of income and the source of expenditure.The heat pump systems,as the main body of building energy,accounts for 15%-20% of the total energy consumption of the society,and its energy saving work is of great significance.With the long-term use of heat pump systems,a large number of performance degradation and increased energy consumption have occurred,causing huge waste of energy.Diagnosing the performance degradation of the heat pump systems timely and maintaining the system's long-term efficient operation is the key way to achieve energy savings in the heat pump.Aiming at the phenomenon of heat pump systems performance degradation,this paper innovatively proposes the concept of sub-health of heat pump systems,and proposes a deep learning-based intelligent diagnostic method to integrate the maintenance of heat pump systems with technologies of the Internet of Things,big data,machine learning,and cloud monitoring,aims to realize the online intelligent monitoring of the health status of the heat pump systems,and further promote the integration of information and intelligence technology in the heat pump field.The main research contents and main contributions of this paper are listed as follows:(1)To clarify the intermediate state of heat pump system health and failure,quantify the energy efficiency change in this state,put forward the concept of sub-health of heat pump system,define the system state that can still complete the established work tasks but increase energy consumption,and use ASHRAE RP-1043 data set and selftest data set to quantitatively analyze the changes of energy consumption,cooling capacity and COP of different sub-health.The experiments respectively verified the sub-health energy efficiency changes of large and medium-sized water cooling systems and small air-cooled systems.The results show that the sub-health of heat pump systems will cause an increase of 2%-12% of energy consumption,and timely diagnosis and reasonable maintenance are of great significance to the energy-saving and efficient operation of the system.(2)In order to solve the problems of difficult modeling of large-lag,unsteady-state systems such as heat pump systems,a deep learning model combining convolutional neural network,codec-decoder,and recurrent neural network is proposed.The model can directly process multi-dimensional time series data and provide benchmark information for sub-health diagnosis methods.The hyperparameters of the benchmark model are studied and optimized,and other four benchmark models has been compared with.The results show that the model proposed in this paper can well fit the system's hysteresis characteristics and greatly improve the modeling accuracy of the unsteady systems,the modeling accuracy is much better than other models.(3)In order to solve the problems of frequent migration of operating conditions,highly complex systems,wide range of sub-health types,and multiple sub-health diagnosis,making full use of the feature extraction capabilities of deep learning models,a sub-health online diagnosis algorithm of heat pump system based on convolutional neural network is proposed.The algorithm combined with the benchmark model can complete diagnosis under unsteady-state conditions such as dynamic changes in the working conditions and environment,which is different from steady-state laboratorylevel diagnosis and achieves online diagnosis.This method reduces the dependence on the prior knowledge of the heat pump system,and uses data driven method to realize the sub-health intelligent diagnosis of the heat pump systems.It does not need to set a judgment threshold and can accurately diagnose concurrent sub-health.The self-built online experimental platform is used to achieve online diagnosis verification,all kinds of sub-health diagnosis accuracy rate exceeds 90%,response time is less than 6 minutes,which proves that the diagnosis method can accurately diagnose under the conditions of large fluctuations in working conditions and environments,and meets the requirements of online diagnosis algorithms.(4)In order to solve the problem of insufficient data and imbalanced data set in online systems,a data augmentation method based on generated adversarial networks is proposed.Using a small number of labeled samples to drive the adversarial training of two neural networks,the quality of data generation is continuously improved by the method of self-game,and finally the generation of high-quality simulation data is achieved.In order to reduce the difficulty of data generation,it is proposed to directly generate sub-health residual data instead of generating original running data,which greatly improves the quality of data generation.In the convolution feature space,the MMD index and the 1-NN index are used to verify the quality of the generated data,and the difference between the generated distribution and the real distribution is analyzed to prove that the generated data is close to the real data distribution.The diagnosis method was used to verify the influence of the generated data on the final diagnosis accuracy,and it was proved that the data augmentation method based on the generated adversarial network can effectively improve the sub-health diagnosis accuracy under the condition of limited real data.This paper focuses on the key problems of online diagnosis,focusing on solving the technical problems include low diagnostic accuracy of unsteady systems,difficulty in multiple sub-health diagnosis,and low diagnostic accuracy under insufficient data set and imbalance data set.After verification by the online experimental system,the method proposed in this paper can be well adapted to the online diagnosis of sub-health of heat pump systems,and the accuracy and real-time performance can meet the requirements accuracy,which indicates that the proposed method has broad application prospects. |
PDF Full Text Request